Rising star in quant finance: David Itkin

Risk Awards 2026: LSE academic spearheaded project that found neat method for calculating price impact of trades

Photo: Juno Snowdon

In quantitative finance, as in other areas of technology, the simplest ideas are often the most effective. So it has proved for David Itkin, this year’s winner of the Risk.net award for rising star in quant finance.

Itkin, an assistant professor in statistics at LSE, has developed a new way of assessing price impact in portfolios of trades. In doing so, he has simplified past methods in a manner that has drawn plaudits from peers.

Johannes Muhle-Karbe, head of mathematical finance at Imperial College London, who collaborated with Itkin on the research, says: “What was most surprising for us is that something so simple actually works so well.”

The germ of the idea came from another of Itkin’s collaborators, Peter Schmidt, a quantitative researcher at Qube Research and Technologies (QRT). While studying for a master’s at Imperial in 2022, Schmidt was exploring strategies for portfolio optimisation. This involved looking at the relationship between price impact and other common factors such as trade signals, risk, and trading costs.

Muhle-Karbe proposed extending the work into a full paper. Itkin, also at Imperial completing his postdoc at the time, joined the project spearheading the theoretical understanding of the research. The ensuing paper, Tackling nonlinear price impact with linear strategies, was submitted in 2023 and published the following year.

You have to choose the right linear strategy for your problem, and this is best done using an optimisation procedure

David Itkin

Price impact occurs when a trade order adversely moves the asset’s price. The dangers of price impact are widely understood by portfolio managers, who must balance the resulting costs against other variables. But the optimal way of modelling this fundamental trade-off is not a perfect science.

Currently, for ease of implementation, portfolio managers assume a linear relationship between the amount of trading and its effect on the asset’s price. Multiplying this relationship by the volume of trades gives what mathematicians describe as an “assumed quadratic” figure for costs.

In reality, though, the relationship is not linear but concave, following square root law. Thus, many firms are unaware how far their own straight-line approximations are from the optimum policy.

“Impact costs are always assumed quadratic in portfolio optimisers. Yet how much is actually lost by doing this approximation had never been clearly answered before,” says Xavier Brokmann, a quantitative researcher at Optiver and co-author of the paper.

Itkin et al’s paper sets out to demonstrate that while naively chosen linear strategies result in performance losses, better performing linear strategies rely on choosing the optimal “effective” quadratic cost parameter.

“You have to choose the right linear strategy for your problem, and this is best done using an optimisation procedure,” explains Itkin.

Portfolio managers have lots of unknowns to worry about, and now this particular one has been taken care of

Xavier Brokmann, Optiver

In earlier versions of the work, Schmidt determined the optimal linear policy by creating a grid of parameters over a long time period and optimising across this grid. Itkin proposed simplifying the framework to a scale optimisation problem of a single function, rather than using a convergence procedure.

The mathematical soundness and interpretability of the proposed method was one of the aspects praised by the award selection committee. Ease and convenience are further benefits of the method: the authors set out to present optimal strategies using infrastructure readily available to practitioners without the need of a computationally expensive algorithm. Muhle-Karbe describes how the theory for linear strategies means calculations can now be made analytically without the need for time-consuming simulations.

Itkin’s postgraduate career includes a stint on the trading floor at the National Bank of Canada, after which he completed a PhD specialising in stochastic portfolio theory at Carnegie Mellon University. He made a transatlantic move to Imperial College London in 2022 to take up a Chapman fellowship position. It was here that he met Muhle-Karbe. “My position was funded by the department, but I was working with Johannes very closely. So essentially, I consider Johannes my postdoc mentor.”

The city offered Itkin the avenues for applied research he was looking for, thus when the paper presented itself, it gave him a chance to align his theoretical expertise with a problem facing the industry.

“If I can get some opportunities to work with people in the industry, I could really learn more about the practical side of things. And that was one of my goals,” he says.

Accuracy

Typically, optimal trading strategies assume that linear price impact corresponds to quadratic costs. However, as Brokmann explains “impact cost is not quadratic. It’s sub quadratic, between linear and quadratic.”

As a result, the paper considers a family of feedback policies with “effective quadratic cost” treated as the tuning parameter alongside a parameter reflecting risk aversion. At a given level of risk, the performance of the linear policy has a fully explicit expression for any value of the effective quadratic cost. This can then be optimised for best performance reflecting realistic nonlinear price impact.

“We set up an optimisation problem where the parameters can be tuned to optimise the new objective function, which accounts for the nonlinear relationship,” says Itkin.

Photo: Juno Snowdon

Their results are benchmarked against the numerical algorithm proposed in a paper by Kolm and Ritter, which optimises trades for multiperiod portfolio selection. Itkin and collaborators compared their framework’s optimal linear policy for a wide range of risk levels against the nonlinear optimiser run at high accuracy. The linear approach resulted in a 2% performance drop versus the nonlinear one.

Given the higher computational costs of the nonlinear approach, many firms may deem a 2% drop by the optimal linear policy to be an acceptable trade-off.

Brokmann says the paper’s findings are reassuring for the industry in that approximations can achieve almost optimal results. “Portfolio managers have lots of unknowns to worry about, and now this particular one has been taken care of,” he says.

“There was no clear indication whether solving the exact optimisation problem would lead to a performance increase of just a few percentage points, or by a factor of three or multiples.”

A follow-up paper is already in the works, says Itkin, incorporating the effects of impact decay. Currently under review, the paper contributes to literature on the nonlinearity of the price impact function incorporating neural network strategies.